Instrument and Witness: A Working Map of Mind and Practice — for the Scientifically Trained

Macheng Shen

Epistemic status: a working synthesis, not a finished theory. I am scientifically trained, and the language I actually think in is the language of dynamics, information, and physics — not Sanskrit or Classical Chinese. So this is my attempt to build a modern-scientific map of consciousness and of contemplative practice, in a language I can reason in rather than only “feel.” It includes its own failure: about halfway through, an adversarial review showed that a good part of my tidy synthesis had become the very thing it was trying to avoid, and I’ve left that in, because the failure is the most useful part. Disclosure, per the norm of this site: this was produced through my agent system and then checked by me. Corrections welcome — the source is on GitHub.

How to read this

Part 1 is a plain-language tour — no equations, just analogies. If you have basic scientific training and have never touched this material, start there and you should be able to reason your way in, not just nod along. Everything after Part 1 is the same ideas with the rigor put back, for anyone who wants it. You can stop after Part 1 and have the whole shape.

I wrote the plain part first and deliberately, because most writing in this area asks you to already believe, or to “get it” by feel. I wanted a version you could follow with the ordinary rational equipment of someone who has taken a physics or a stats class.

Part 1 — In plain language

1. The one question

Is consciousness a light switch or a dimmer?

Switch: you either have it or you don’t, and somewhere in evolution humans (or maybe mammals) flipped it on. Dimmer: it’s a continuous quantity — near-zero in a thermostat, low in an insect, higher in a dog, high in us — and nothing ever flipped, some systems just turned it up.

That’s the whole question, and how you answer it changes what you think you are.

2. Two layers: the movie and the screen

Watch any moment of experience closely and it splits in two.

There’s the content — thoughts, images, sensations, emotions, the ongoing “movie.” And there’s the awareness the movie is showing up in — call it the screen, or the light the movie is projected with.

Here’s the part that takes a second to land: the “self” you normally defend — my name, my story, my personality, my plans — is a character in the movie. It’s content. The awareness is not a character; it’s what the movie is playing on.

The striking thing, and the reason this is worth taking seriously rather than dismissing as mysticism, is that wildly independent traditions — Buddhist, Hindu, Daoist, Greek, Jewish — each found this same split on their own and each said the same warning: you keep mistaking the character for the screen.

Modern brain science draws almost exactly the same line, just with different names. It can explain the movie — how the brain integrates information, directs attention, builds a model of itself, reports on its own states. It cannot explain the screen — why there’s any “showing up” at all, why all that information-processing is lit rather than happening in the dark. That gap has a name (the “hard problem”), and it is unsolved by everyone, including me.

3. The answer, in plain terms

Split the question along that seam and it resolves into two very different answers:

• The screen (bare awareness) is probably not special to humans. More dimmer than switch. This is the honest bet of most of the traditions and of the more serious modern “consciousness is fundamental” positions — but it is a bet, not a proven fact, because nobody has closed the question of why there’s a screen at all.
• What’s special about humans lives in the movie. Our movie contains a character that can watch itself, notice its own patterns, and rewrite its own script — and, apparently uniquely, hand that script to the next generation through language and culture. That’s the real human distinctive: not a rarer kind of awareness, but a movie sophisticated enough to edit itself and be copied forward.

So: awareness, probably everywhere and graded; human specialness, a structural achievement inside the content, not a different substance of consciousness.

4. How the past grips the present: mud vs spring

A short detour that turns out to matter. History can hold onto a system in two very different ways.

• Mud. A drag that resists motion in proportion to how fast you’re going. It slows you, it forgets, it never gives anything back.
• Spring. It stores what you push into it and gives it back later. It can overshoot, oscillate, carry momentum.

A habit modeled as mud is “feedback” — a bias that fades. A habit modeled as a spring is “inertia” — stored momentum that carries you even when you’d rather stop. From the inside they feel similar. They are mathematically different, and it turns out most of what we casually call habit or “karma” is mud, not spring — a fading drag, not real stored momentum. (I worked this out carefully in a separate short physics piece — the one line that survives is: forgetting is not momentum; to have real momentum, memory has to store and return.)

Why does this belong here? Because it’s a clean, small example of the whole method: take a vague intuition (“the past weighs on me like inertia”), put it in honest modern-scientific form, and find out it was half right and exactly which half. That is what I want from a framework.

5. The trap: mistaking the map for the territory

There’s a failure mode with an old name — 所知障, roughly “the obstruction of the knowable”: when your knowledge about a thing piles up until it blocks direct contact with it. You can assemble a perfect library on swimming and never once get wet.

I hit this head-on. When I tried to unify all these traditions plus modern science into one clean scientific map, I ran an adversarial review against my own synthesis — and it showed that the map itself had become an instance of the trap. My tidy “movie vs screen” grid was partly something I had imposed: several traditions actually reject the split (some say both layers are equally empty; some say the “screen” is itself active, not a passive witness). Counting them all as agreeing was me finding my own assumption everywhere I looked. And the elaboration — the ever-neater table — had quietly started substituting for the looking.

The honest lesson, and the reason I’m keeping the failure in public: a map of mind is a raft to cross on, not a monument to defend. Cut the map, keep the practice.

6. So is a scientific framework even worth building? Yes — if the authority points the right way

Given all that, why build a framework at all? Because a raft is genuinely useful — if you keep the authority pointed correctly.

Simple test: when your framework and your direct experience disagree, which one wins? If you’d trust the map over your own feet, you’re lost — the map has become a cage. If experience always gets the final vote, and the theory is only there to tell you where to look next, the theory is a good raft.

And building it in modern-scientific language specifically is not a betrayal of the traditions — it’s the same move they made. Their maps (the Buddhist eight-consciousness model, the yoga-sutra model of mind, and so on) were built in each culture’s sharpest available language to guide practitioners. Modern-scientific language is simply the sharpest one I actually own. There’s one extra hazard, though, and it’s worth naming: scientific language loves to treat its models as the real thing. It has no built-in “now drop the map” step the way the contemplative traditions do (they end with “and this teaching too is a raft”). So if you build in this language, you have to add the drop-the-map discipline by hand.

7. How to actually start looking

You can’t get usable data from a shaky camera. The first job of practice isn’t insight — it’s stabilizing the instrument: training attention until it’s steady enough to watch experience without stirring it up. (This is what the “just watch the breath” starter instruction is quietly for.)

Then a genuinely rational method opens up. Instead of asking “what was I thinking about” (content), you ask “how did that thought arise — what came just before it, what was its texture, where did it go” (structure). You replay a single moment in slow motion.

And there’s a whole small science built on exactly this — neurophenomenology — which pairs these trained first-person reports with brain measurements so each corrects the other. It’s the one part of this entire enterprise that is honestly, third-person testable. If you want a doorway into all of this that stays rational and low on mysticism, that’s the one I’d point to: train the camera, then trust the footage, then check it against the instruments. The rest of this piece builds out each of these threads.

Part 2 — The physics leg: memory, inertia, feedback

The “mud vs spring” idea, done properly, is a question about memory kernels — how a system’s past enters its present through a weighted integral over its own history. Two intuitions compete for what such a memory is: control-theoretic feedback (the past sets a gain) or Newtonian inertia (the past gives you mass). They are usually treated as interchangeable. They aren’t.

The result, in one paragraph: a purely dissipative, forgetting memory (a positive kernel that just decays) produces, at low frequency, an effective mass with the wrong sign — it cannot give you stable inertia. To get genuine inertia the memory has to have a reactive, storing component — formally, its velocity kernel must have a negative lobe, which is what a hidden mode that stores-and-returns looks like once you integrate it out. And the induced “mass” vanishes when that hidden mode becomes very fast, leaving only feedback. So inertia and feedback are the reactive and the dissipative parts of one memory — and modeling any history-dependence (a habit, an accumulated bias) as “inertia” quietly assumes a storage mechanism that plain forgetting does not have.

The full derivation — with the frequency-domain algebra, a reproducible numpy toy, and (importantly) a public list of the corrections a technical review sent back — is a separate piece:

When Does Memory Become Mass? Inertia and Feedback in Systems with History

I flag it here as a worked example of the whole stance: an intuition, made honest, turns out half-right, and you learn precisely which half.

Part 3 — The consciousness question, with the rigor put back

The plain “movie vs screen” split is, more carefully, a split between two layers of mind:

• an instrument layer — structure, dynamics, memory, an update rule; the part that processes and models, including modeling itself; and
• a witness layer — the bare knowing in which the instrument’s activity appears.

My working theory of the instrument layer is an ordinary dynamical one: mind as an embedded, memory-carrying control loop that builds a model of its world and of itself as an agent in that world, and updates that model to reduce surprise. In the current science this is essentially Active Inference / the Free Energy Principle — a self-evidencing agent. It is a theory of the instrument, and it is silent about the witness. That silence is honest, not a gap I’m hiding: it’s the hard problem, conceded up front.

The decomposed answer to “special or universal”:

• Witness / phenomenality: most contemplative traditions, and the more serious modern “consciousness is fundamental” positions (panpsychism, and process philosophy in the West), vote universal and graded. If phenomenality is fundamental, it is not what makes humans special. This is a metaphysical bet, and it is not currently falsifiable — I label it as such.
• Instrument / structure: here specialness is real but graded, and empirical. What is plausibly human-distinctive is a depth of recursive self-modeling (a loop that can represent and edit its own update rule) plus a symbolic / cultural transmission channel that carries the instrument’s contents beyond a single body and a single life. Recursion and cumulative culture are the strongest human-distinctive candidates in the empirical literature; basic sentience and valence appear widely shared.

So: awareness plausibly universal; human specialness a structural achievement in the instrument, not a different witness.

The traditions, used as a quarry (not as authorities)

I read the traditions as raw material for my own synthesis, not as scripture. Compactly, and honestly about where they map and where they don’t:

Two families of genuine agreement are worth more than the rest:

• The update loop. The Yogācāra “seeds ripen into experience, experience re-perfumes the seeds,” and the Yoga-sutra “each mental modification leaves a groove that biases the next,” are close in shape to a history-carrying update operator. (Close in shape — not a precise mapping; the resemblance is at the level of “history biases the present,” which is generic. Don’t oversell it.)
• The witness/instrument seam itself, found independently again and again.

And the divergences I will not paper over — because forcing them to agree is exactly the trap of Part 1 §5:

• Opposite goals. My framework wants to grow and refine the instrument (a better model, a lighter agent). Buddhism and Daoism say the highest move is to see through / un-build it. Same machinery, opposite direction. I don’t get to dissolve that with a clever reframing.
• The witness’s texture is disputed even among the traditions — empty mirror vs full self-luminous presence. My framework has nothing to say, and shouldn’t pretend to.
• Cessation and timelessness (a state with no next moment) have no analog in any dynamical theory at all.

Part 4 — Where the map broke (and why I’m keeping it public)

I ran an adversarial review against the synthesis in Part 3. It landed several hits I’ve accepted:

1. The instrument/witness split is partly my projection, not a universal discovery. At least two traditions above actively refuse it (Madhyamaka: both layers are empty, so there’s no witness-residue to file; Kashmir Śaivism: the “witness” is intrinsically active, which collapses the wall). I had bent them to fit my grid.
2. “Independent convergence” was oversold. Some of those traditions share direct historical lineage (Neoplatonism → Kabbalah), so they aren’t independent votes; and an analyst who already holds the instrument/witness distinction will find it everywhere. Convergence is real but weaker than “everyone agrees.”
3. My own theory over-claims by breadth. Saying it “cleanly owns the instrument layer” sounds strong, but a framework loose enough to rhyme with every tradition’s language is loose because it’s near-unfalsifiable — the breadth is a symptom, not a triumph. Active Inference has exactly this liability: it describes almost any adaptive system.
4. Relabeling doesn’t dissolve the hard sub-problem. Even granting “the instrument is a generator, not the self,” you still owe an account of what binds the instrument’s many parts into one apparent point of view — and the traditions give opposite answers.

The deepest version of the critique: the whole activity of tabulating and unifying is itself a candidate for 所知障 — subtraction-work (see less, cling less) misfiled as addition-work (build a bigger, prettier model). The tell was subtle and worth stating: my guard against over-unifying had itself become a proud feature of the framework. The reflex was congratulating itself for resisting the reflex.

What survives the razor:

• the instrument-layer theory, held honestly as a near-unfalsifiable organizing frame whose real, defensible win is the meta-problem (explaining why a system would represent itself as having ineffable experience — not explaining the experience);
• the genuine, un-unifiable divergences (they’re informative because they don’t merge);
• and one actionable residue — the first-person practice, and its testable pairing with measurement.

Cut the map, keep the practice. The map was the scaffold that got me to that sentence; now it comes down.

Part 5 — Building your own scientific framework: valid, or self-obstruction?

The natural worry, after Part 4: if map-building tips so easily into 所知障, is it a mistake to build a personal modern-scientific framework for this at all?

No — and the discriminator is a single variable: which way authority flows between the map and the practice.

• Framework serves practice (experience is the final court; when the map and the sitting disagree, the map yields; the theory is a raft) → valid.
• Practice serves framework (the theory becomes the goal, and is defended against experience) → 所知障.

Same activity; the direction decides whether it’s medicine or disease. Three field-tests you can run on yourself in real time:

1. The raft test. Next time direct experience contradicts your framework, do you drop the framework or explain the experience away? Willing to drop it → medicine. Defending it → disease.
2. The direction test. Is refining the framework sending you to the practice, or replacing it? If model-building is reducing your actual practice time, it has flipped.
3. The avoidance test (the subtle one). “I understand it conceptually” is an easy hiding place from the discomfort of actually looking. Theory-building can be avoidance dressed as preparation.

And the language-specific caution, stated once more because it’s the crux: the old maps were built by practitioners, from first-person data, and came with an explicit “drop this too.” Modern-scientific language is third-person and objectifying, and has no built-in drop-the-map move — it tends to reify the model into the thing. So a scientific framework for practice is valid, but carries a higher reification risk, and you have to import the raft-discipline manually. (Part 4 is a live specimen of the risk: an anti-over-reach guard that mutated into a trophy.)

Part 6 — The practice bridge: neurophenomenology

This is where the whole thing becomes something you can do and, in principle, test.

The core idea (Francisco Varela’s, 1996): pair a rigorous first-person method with third-person brain measurement, and let each constrain the other. The direction of constraint is the point, and it’s the same “authority flows practice→theory” from Part 5: first-person data disciplines the model; the model only points back at where to look.

Why “trained” first-person reports. Naïve introspection is unreliable — a well-known result in psychology is that people confidently report more than they can actually know about their own processing. The fix is not to abandon first-person data but to train the instrument: stabilize attention (the classic concentration practices) until reports become steady, fine-grained, and repeatable. This is exactly what the contemplative methods are for, read as instrument-calibration rather than doctrine.

A method you can use solo (micro-phenomenology, developed by Pierre Vermersch and Claire Petitmengin): after an experience, don’t ask what (content), ask how — how did this arise, what immediately preceded it, what was its texture, how did it pass? You guide attention back to the pre-reflective structure of a single moment. It’s a repeatable protocol, not a mood.

What to attend to — the vocabulary trained practitioners learn to report, i.e. the actual targets:

• the arising and passing of moments (their impermanence, watched directly, not believed);
• the gap between two thoughts;
• the difference between being caught in an experience and witnessing it;
• the how / texture of an experience rather than its content.

These are the instrument/witness seam from Parts 1 and 3, turned into concrete first-person targets you can check in your own experience — which is what makes this the falsifiable-in-practice end of the whole subject.

That it’s real science, not hand-waving — three landmark results:

• first-person reports of a subject’s own preparatory state were used to make otherwise-noisy EEG dynamics interpretable (Lutz, Lachaux, Martinerie & Varela, 2002);
• first-person descriptions of the aura preceding epileptic seizures were used to anticipate them (Le Van Quyen & Petitmengin, 2002);
• long-term meditators self-induce high-amplitude gamma synchrony during practice, giving a unique window on trained states (Lutz, Greischar, Rawlings, Ricard & Davidson, 2004).

If you want one rational, low-mysticism doorway into everything above: train the camera, then trust the footage, then check it against the instruments.

Part 7 — Honest status and open problems

• The instrument-layer theory is a decent organizing frame, essentially today’s best science (Active Inference), and by that same token near-unfalsifiable as stated. Its honest achievement is the meta-problem, not the hard problem.
• The binding / combination problem — what makes many parts into one point of view — is unsolved here, and shows up in the same shape in panpsychism and in my “nested self” picture. I don’t have it.
• The grow-vs-see-through split (refine the instrument vs. dissolve it) is a real fork I haven’t resolved and may not be resolvable by theory at all — it may be a choice, not a fact.
• The one part with a path to being testable is the neurophenomenology program. That’s where I think the actual work is, and where a framework earns its keep instead of accreting.

If this mapped a small piece of the territory for you — especially the plain-language part — that’s the point. It’s a raft. Use it to get somewhere, then put it down.

References and doorways

Physics leg: When Does Memory Become Mass? (this site).

Science of consciousness: Chalmers, “Facing Up to the Problem of Consciousness” (1995) and “The Meta-Problem of Consciousness” (2018); Friston, the Free Energy Principle / Active Inference (Parr, Pezzulo & Friston, 2022); Dehaene, Consciousness and the Brain (2014); Frankish, “Illusionism as a Theory of Consciousness” (2016); Graziano, Rethinking Consciousness (2019).

Neurophenomenology / first-person method: Varela, “Neurophenomenology” (1996); Varela, Thompson & Rosch, The Embodied Mind; Lutz & Thompson, “Neurophenomenology” (2003); Petitmengin, “Describing one’s subjective experience in the second person” (2006); Evan Thompson, Waking, Dreaming, Being (2015).

Traditions (primary doorways, as a quarry): the Cheng Weishi Lun and Vasubandhu’s Triṃśikā (Yogācāra); the Māṇḍūkya Upaniṣad with Gauḍapāda’s Kārikā, and Patañjali’s Yoga Sūtras I.2 (witness / mind-as-modifications); Zhuangzi, the “fasting of the heart” and “sitting-forgetting” passages; the Spanda Kārikās and Pratyabhijñāhṛdayam (Kashmir Śaivism); Plotinus, Enneads V (emanation).

The method itself: build the map in the language you can actually reason in; keep experience as the final court; and remember the map is a raft.